\documentclass[11pt]{article} \usepackage[T1]{fontenc} \usepackage{lmodern,amsmath,amssymb,amsthm,microtype} \usepackage[margin=1in]{geometry} \usepackage{booktabs,graphicx,tabularx,placeins} \usepackage[table]{xcolor} \definecolor{proved}{HTML}{E4F0E8} \usepackage[colorlinks=true,linkcolor=black,citecolor=black,urlcolor=blue]{hyperref} \newtheorem{theorem}{Theorem} \newcommand{\R}{\mathbb R} \newcommand{\vol}{\operatorname{vol}_d} \title{Erd\H{o}s's 1974 distance-product volume problem resolved: the ball as unique maximizer, sharp inradius stability, and other miscellanea} \author{Dave Goel} \date{\small First version: November 19, 2026. This version: November 19, 2026.} \begin{document} \maketitle \begin{abstract} We consider the set of all points of $\R^d$ whose distances to $n$ given points have product at most one. Erd\H{o}s (1974) asks how large its volume can be. We show that in every dimension $d\ge3$, and for every integer $n\ge1$, this volume is at most that of the unit ball. Equality holds only and exactly when all $n$ given points coincide. This completely answers Erd\H{o}s's question; in the plane, the same statement is P\'olya's theorem of 1928. In fact we settle a more general problem, with weighted points. Our theorem also shows that the two-point counterexample in dimension three, reported by Erd\H{o}s in 1976 and repeated in 2018 and 2026, cannot exist. We also prove a sharp stability estimate: for each fixed $d\ge3$, if such a set falls short of the maximal volume by a small fraction $D$, it still contains a ball of radius at least $1-(1/(d-2)+o(1))D$ as $D$ tends to zero, uniformly in the number, weights and positions of the points. The same holds for every compactly supported probability measure in place of the points. Two equally weighted points approaching each other show that the coefficient $1/(d-2)$ cannot be lowered. Finally, the product of the distances is at most one exactly when their geometric mean is at most one, and we ask what happens when the geometric mean is replaced by the weighted power mean of order $p$. We show that the same volume bound holds for all placements exactly when $p\ge2-d$, and that in this range equality again requires coincident points. For $p\ge2-d$ we deduce the bound from the weak estimate of Tang and Wang; for $p<2-d$ the polarization theorem of Erd\'elyi and Saff gives placements that violate it. We prove the equality statement at the endpoint $p=2-d$ directly, and for $p<2-d$ we compute the leading term of the excess volume of the region of two nearby points. \end{abstract} \noindent\textit{This is the expanded version of the paper \textquotedblleft{}Erd\H{o}s's 1974 distance-product volume problem resolved: the ball is always the unique maximizer in $d\ge2$\textquotedblright{}~\cite{short}.} \section*{The volume problem} Fix a dimension $d\ge2$ and a number $n\ge1$. Place $n$ points $a_1,\ldots,a_n$, called marks, anywhere in $\R^d$; marks may coincide. Let $E$ consist of the points $x$ for which $|x-a_1|\cdots|x-a_n|\le1$. \emph{How large can the volume of $E$ be, and which placements of the $n$ marks attain the maximum?} Erd\H{o}s's question dates from 1974; it is Problem~4.23 in the 2018 list of Hayman and Lingham~\cite[Problem~4.23, Tables~1--2]{hayman}. If all $n$ marks lie at one point $a$, the product is $|x-a|^n$, and $E$ is the closed unit ball about $a$. We write $\omega_d$ for its volume; thus $\omega_3=4\pi/3$. We prove that, in every dimension $d\ge3$ and for every $n$, the largest volume of $E$ is $\omega_d$, and that every placement in which the marks do not all coincide gives strictly less volume (Theorem~\ref{thm:volume}). In fact we settle a more general problem: Theorem~\ref{thm:volume} allows the marks to carry different weights. Erd\H{o}s's problem is the case of equal weights. \begin{center} \small \begin{tabularx}{\linewidth}{@{}XX@{}} \toprule \textbf{Best result before this paper} & \textbf{New result}\\ \midrule \multicolumn{2}{@{}p{\linewidth}@{}}{\emph{What is the largest volume of $E$, for $n$ marks in dimension $d\ge2$, and when is it attained?}}\\[.3em] Known only for $d=2$ (P\'olya~\cite{polya}). Listed as open for $d\ge3$~\cite[Problem~4.23]{hayman}, \cite[abstract]{ferudun}. & Fully resolved for all $d\ge2$ and $n\ge1$: the maximum is $\omega_d$, attained exactly when all marks coincide.\\[.4em] \midrule \multicolumn{2}{@{}p{\linewidth}@{}}{\emph{Can two distinct marks in dimension three give more volume than the unit ball?}}\\[.3em] Reported possible, incorrectly. Erd\H{o}s wrote in 1976 that Piranian had a counterexample~\cite{erdos}. Hayman and Lingham repeated this in 2018~\cite{hayman}, and Ferudun in 2026~\cite{ferudun}. & No. Every pair of distinct marks gives volume strictly less than $4\pi/3$. The reported counterexample does not exist.\\ \bottomrule \end{tabularx} \end{center} \begin{figure}[!ht] \centering \includegraphics[width=.77\linewidth]{figures/two_regions.pdf} \par\smallskip {\small\begin{tabularx}{\linewidth}{@{}l>{\raggedright\arraybackslash}X>{\raggedright\arraybackslash}X@{}} \toprule & The two marks coincide & The two marks are distinct\\ \midrule Earlier status & \cellcolor[HTML]{E5E6E7}Ball volume $4\pi/3$ known; is it maximal? & \cellcolor[HTML]{E5E6E7}A larger volume was reported, incorrectly, for a distinct pair\\ Theorem~\ref{thm:volume} & \cellcolor[HTML]{D8EBDD}Yes: maximum volume $4\pi/3$ & \cellcolor[HTML]{F5DBD7}No: every distinct pair gives less\\ \bottomrule \end{tabularx}} \caption{Two marks in dimension three, drawn to the same scale. Left: the marks coincide, and the region is the unit ball. Right: the marks are $2.5$ units apart, and the region splits into two pieces whose combined volume is less than that of the unit ball. The separation $2.5$ is our choice; Erd\H{o}s's report gave no separation.}\label{fig:regions} \end{figure} \FloatBarrier To see how the dimension enters the problem, take two marks at distance $2t$ from each other, with $01$ the region splits into two intervals. Their total length is \[ 2\bigl(\sqrt{t^2+1}-\sqrt{t^2-1}\bigr) =\frac{4}{\sqrt{t^2+1}+\sqrt{t^2-1}}<2\sqrt2. \] So separating the marks by more than two units makes the total length of the region smaller. \item \emph{The plane, $d=2$.} The unit disc is the best region. P\'olya proved in 1928 that, for every $n$ and every placement of the marks, the area of $E$ is at most $\pi$, with equality only when all the marks coincide~\cite[Satz~II, p.~280]{polya}. \item \emph{Every dimension $d\ge3$.} The unit ball is the best region: by Theorem~\ref{thm:volume}, the volume of $E$ is at most $\omega_d$, with equality only when all the marks coincide. \end{itemize} In three dimensions, Erd\H{o}s at first expected the unit ball to have the largest volume~\cite[\S4]{erdos}. A single mark always gives a unit ball, so two marks are the first case in which the placement matters. In 1976, Erd\H{o}s reported that Piranian had found a two-point counterexample to the claim that the unit ball maximizes volume in three dimensions~\cite[\S4, pp.~350--351]{erdos}. Erd\H{o}s gave neither the configuration nor a volume calculation. That report is incorrect: the counterexample cannot possibly exist. In the proof of Theorem~\ref{thm:volume}, we show that every pair of distinct marks gives volume strictly less than $4\pi/3$, the volume attained when the two marks coincide. We know of no publication before the present paper that explicitly corrected Erd\H{o}s's report. Hayman and Lingham repeated the report in 2018~\cite[Problem~4.23, p.~79]{hayman}. Ferudun repeated it in 2026, in a summary of Hayman and Lingham's Problem~4.23~\cite[\S1.1, p.~1]{ferudun}. Problem~4.23 asks a second question as well: whether $E$, when it is connected, lies in a ball of radius two. Ferudun answered that covering question, in every dimension~\cite[Theorem~1.1]{ferudun}. The present paper concerns the volume question only. We state Theorem~\ref{thm:volume} in the more general form, with weighted marks. Give each mark $a_j$ a weight $w_j>0$, with the weights summing to one. The product of the distances is replaced by their weighted geometric mean $G$, which is compared with an arbitrary threshold $R>0$. Erd\H{o}s's region is the case of equal weights $w_j=1/n$ and $R=1$, because the product of $n$ distances is at most one exactly when their geometric mean is at most one. Thus $E=E_1$ in the notation of the theorem. \begin{samepage} \begin{theorem}\label{thm:volume} Let $d\ge3$, let $a_1,\ldots,a_n\in\R^d$ with $n\ge1$, and let $w_1,\ldots,w_n>0$ have sum one. For $R>0$ put \[ G(x)=\prod_{j=1}^n |x-a_j|^{w_j},\qquad E_R=\{x:G(x)\le R\}. \] Write $\omega_d$ for the volume of the unit ball in $\R^d$. Then $\vol(E_R)\le \omega_d R^d$, with equality exactly when the $a_j$ coincide. In particular, for $d=3$, $w_j=1/n$ and $R=1$, the maximum in Erd\H{o}s's problem is $4\pi/3$. \end{theorem} \end{samepage} For equal weights and $R=1$, Theorem~\ref{thm:volume} answers Erd\H{o}s's volume question in every dimension $d\ge3$: the largest volume is $\omega_d$, and it is attained exactly when all marks lie at one point. We first treat two marks in dimension three, the case of the reported counterexample, by applying P\'olya's planar theorem to each cross-section. We then give two proofs of Theorem~\ref{thm:volume}: a direct proof by flux across level surfaces, which follows Talenti's rearrangement method~\cite{talenti}, and a second proof that combines electrostatics and isoperimetric geometry. The later sections treat the largest inscribed ball, first in dimension three and then in every dimension $d\ge3$. The final section replaces the geometric mean of the distances by a power mean. \subsection*{Two marks in dimension three: a short proof} Let the two marks be distinct, and choose coordinates in which both lie in the horizontal plane of height zero. Take a point at height $h$, and let $u$ and $v$ be the distances from its vertical projection onto the horizontal plane to the two marks. The distances from the point itself to the marks are $\sqrt{u^2+h^2}$ and $\sqrt{v^2+h^2}$, and the square of their product is \[ (u^2+h^2)(v^2+h^2)=(uv+h^2)^2+h^2(u-v)^2. \] For a point of the region the left side is at most one. The right side is at least $(uv+h^2)^2$, so $uv+h^2\le1$, that is, $uv\le1-h^2$. Hence the region has no point at a height $h$ with $|h|>1$, and for $|h|<1$ its horizontal slice at height $h$ lies in the planar region $uv\le1-h^2$. Identify the horizontal plane with the complex plane, and let $F$ be the monic quadratic polynomial whose zeros are the two marks. If $z$ is the vertical projection of the point, read as a complex number, then $uv=|F(z)|$. P\'olya's theorem~\cite[Satz~II, p.~280]{polya} states that, for a monic polynomial $P$ of degree $n$ and every $R>0$, the region $|P(z)|\le R$ has area at most $\pi R^{2/n}$, with equality only when $P(z)=(z-c)^n$ for some complex number $c$. (In this statement $n$ is the degree and $R$ is a bound on $|P|$; $R$ is not the threshold of Theorem~\ref{thm:volume}.) For $P=F$, with $n=2$ and $R=1-h^2$, the planar region $uv\le1-h^2$ has area at most $\pi(1-h^2)$, and strictly less because the two marks are distinct. P\'olya's bound holds for the whole planar region, including both pieces if the region is disconnected. It therefore holds for the whole slice. The slice of the unit ball at the same height has area exactly $\pi(1-h^2)$. Integrating the slice areas over $-11$. This plane separates the two halves. \subsection*{First proof: flux across level surfaces} The idea is to enclose $E_R$ in a larger set, a superlevel set of the Newtonian potential of the marks, and to show that the volume of this larger set is at most $\omega_dR^d$. When two marks lie at different points, the larger set contains an open ball that $E_R$ misses, and this gives the strict inequality. At each mark $a_j$ place a point charge of size $w_j$; from now on we also call the marks \emph{charges}. Write $U(x)=\sum_j w_j|x-a_j|^{2-d}$, and set $U=+\infty$ at each charge. The arithmetic--geometric mean inequality gives $U(x)\ge G(x)^{2-d}$ away from the charges. At a point of $E_R$ that is not a charge we have $G\le R$, and the exponent $2-d$ is negative, so $U\ge G^{2-d}\ge R^{2-d}$. At a charge $U$ is infinite. Hence $E_R\subset\{U\ge R^{2-d}\}$. For $\lambda>0$ let $W(\lambda)=\vol\{U\ge\lambda\}$ and define $r(\lambda)>0$ by $W(\lambda)=\omega_d r(\lambda)^d$. Thus $r(\lambda)$ is the radius of the ball that has the same volume as the superlevel set $\{U\ge\lambda\}$. Each superlevel set is bounded and contains a neighborhood of each charge. At a regular value $\lambda$, let $A(\lambda)$ denote the area of its boundary, the level surface $\{U=\lambda\}$. Because $U$ is harmonic away from the charges, \[ \int_{U=\lambda}|\nabla U|\,dS=d(d-2)\omega_d. \] Indeed, merge charges that sit at the same point, and remove from the superlevel set a small ball about each of the resulting charges. On the level surface the outward normal derivative of $U$ is $-|\nabla U|$. Through each small sphere, in the direction of the normal that points toward its center, the term of $U$ that is singular at the center has flux $d(d-2)\omega_d$ times the size of the charge at that center, and every other term has flux zero. The sizes of the charges sum to one, so the divergence theorem gives the identity. For $0<\kappa\le\lambda<\infty$, call the set where $\kappa\le U\le\lambda$ a \emph{slab}. On every slab, $U$ is smooth and its critical set has volume zero. To see this, note that $U$ is analytic and nonconstant on the connected complement of the charges, so at least one of its partial derivatives is not identically zero there. That derivative is again analytic, so its zero set has volume zero; and this zero set contains the critical set. The same argument applied to $U-\lambda$ shows that each level set $\{U=\lambda\}$ has volume zero. The coarea formula on the slab, applied away from the critical set, shows that $W$ is locally absolutely continuous and that $-W'(\lambda)=\int_{U=\lambda}|\nabla U|^{-1}\,dS$ for almost every $\lambda$. By Sard's theorem, almost every $\lambda$ is a regular value. For almost every $\lambda$ we therefore have \[ (d\omega_d r(\lambda)^{d-1})^2\le A(\lambda)^2 \le d(d-2)\omega_d\,(-W'(\lambda)). \] The first inequality is the isoperimetric inequality for the set $\{U\ge\lambda\}$, whose volume is $\omega_d r(\lambda)^d$. The second is the Cauchy--Schwarz inequality $A(\lambda)^2\le\int_{U=\lambda}|\nabla U|\,dS\cdot\int_{U=\lambda}|\nabla U|^{-1}\,dS$, combined with the flux identity and the formula for $-W'(\lambda)$. Substituting $W=\omega_d r^d$, so that $-W'=-d\omega_d r^{d-1}r'$, and cancelling gives $-(d-2)r^{1-d}r'\ge1$, that is, $(r^{2-d})'\ge1$ almost everywhere. As $\lambda$ decreases to zero the superlevel sets exhaust $\R^d$, so $r(\lambda)^{2-d}\to0$. Since $W$ is locally absolutely continuous, so is $r^{2-d}$, and integrating from zero to $\lambda$ yields $r(\lambda)^{2-d}\ge\lambda$. At $\lambda=R^{2-d}$ it follows that $r(\lambda)\le R$. Since $E_R$ lies in $\{U\ge R^{2-d}\}$, this proves the volume bound. The same bound on the superlevel set follows from Tang and Wang's sharp weak estimate for densities~\cite[Theorem~1.1, $s=2$]{tang} by smoothing the point charges and passing to the limit; see the deduction after Theorem~\ref{thm:power}. It remains to treat equality. Suppose first that two marks lie at different points $a$ and $b$. Consider the ray that starts at $a$ and points directly away from $b$. At every point of this ray the distances to $a$ and to $b$ are different. The compact set $E_R$ contains $a$. Let $y$ be the last point of the ray that belongs to $E_R$. By continuity, $G(y)=R$. Because the distances from $y$ to $a$ and to $b$ differ, the arithmetic--geometric mean inequality is strict at $y$, so $U(y)>R^{2-d}$. A point on the ray just beyond $y$ therefore has both $G>R$ and $U>R^{2-d}$, and by continuity both inequalities hold on a small open ball about it. This ball lies in the superlevel set but outside $E_R$. Thus $E_R$ has strictly smaller volume than the superlevel set $\{U\ge R^{2-d}\}$, and hence volume strictly less than $\omega_dR^d$. If instead all the marks coincide at $a$, then $G(x)=|x-a|$ and $E_R$ is the ball of radius $R$ about $a$, of volume exactly $\omega_dR^d$. \qed \subsection*{Second proof: combining electrostatics and isoperimetric geometry} We treat $E_R$ and a ball of radius $R$ as conductors. Using Clark and Laugesen's logarithmic moment inequality, we show that, when both conductors are held at unit potential, $E_R$ carries at most as much charge as the ball. Newtonian capacity measures the charge that a conductor carries at unit potential. By the isocapacitary inequality, a region with at most the ball's capacity also has at most the ball's volume. The compact set $E_R$ contains a ball and hence has positive Newtonian capacity. Let $c$ be the radius of a ball of the same capacity and let $\nu$ be the Newtonian equilibrium probability measure of $E_R$, the distribution of one unit of charge on $E_R$ that minimizes its Newtonian energy. For a ball, this equilibrium measure is the uniform measure on its boundary sphere. Applying Theorem~1.1(ii) of Clark and Laugesen~\cite{clark}, with the origin moved to $a_j$, gives $\int\log |x-a_j|\,d\nu(x)\ge\log c$. Multiplying by the weights $w_j$ and summing over $j$ gives \[ \log c \le \int_{E_R}\log G(x)\,d\nu(x) \le \log R. \] The second inequality holds because $G\le R$ on $E_R$. The classical isocapacitary inequality (see~\cite[\S1]{capacity}) gives $\vol(E_R)\le\omega_d c^d$. That inequality is stated for bounded open sets. For a compact set, apply it to bounded open neighborhoods of the set and take the infimum of their capacities. Since Newtonian capacity is outer regular, this infimum is the capacity of the compact set. We use the resulting inequality for compact sets here and in the inradius argument below. Since $c\le R$, the isocapacitary inequality gives the volume bound $\vol(E_R)\le\omega_dR^d$. For the equality case, suppose $\vol(E_R)=\omega_d R^d$. Then $c=R$, so both inequalities in the last display are equalities. Each inequality $\int\log|x-a_j|\,d\nu(x)\ge\log c$ enters the sum with a positive weight, so each of them is an equality as well. By Clark and Laugesen~\cite[Theorem~1.1(iv)]{clark}, $E_R$ contains the entire sphere of radius $R$ about each $a_j$. Choose a unit vector $u$, and among the marks choose one, $a$, for which $u\cdot a$ is largest. The point $a+Ru$ is at distance $R$ from $a$. For every mark $b$ that lies at a different point we have $u\cdot(a-b)\ge0$, and therefore \[ |a+Ru-b|^2=R^2+2R\,u\cdot(a-b)+|a-b|^2>R^2 . \] So if two marks lay at different points, the geometric mean $G(a+Ru)$ would exceed $R$. This contradicts the fact that $E_R$ contains the sphere of radius $R$ about $a$. \qed \input{inradius} \input{general_stability} \input{power_means} \section*{Discussion} Extending P\'olya's planar theorem, we proved that in every dimension $d\ge3$ the largest possible volume of the region is that of the unit ball, attained only when all marks coincide. We generalized this result to weighted marks. Our proof corrects Erd\H{o}s's 1976 report of a two-point counterexample in three dimensions by showing that every pair of distinct marks gives strictly less volume than the unit ball. In every dimension $d\ge3$ we also showed that, as the volume approaches its maximum, the worst possible fractional loss of inradius agrees to first order with $1/(d-2)$ times the fractional loss of volume; in three dimensions this coefficient is one. Finally, we replaced the geometric mean by the weighted power mean of order $p$ of the distances, and showed that the volume bound then holds for all marks and weights exactly when $p\ge2-d$. These results suggest several further questions. First, fix a dimension $d\ge3$ and let $\varepsilon$ tend to zero. Among regions at threshold one with at least $(1-\varepsilon)$ times the volume of the unit ball, the sharp lower bound for the inradius is $1-\varepsilon/(d-2)+o(\varepsilon)$. This is Theorem~\ref{thm:radius} when $d=3$ and Theorem~\ref{thm:all-dimensions} in general. What is the next term of this expansion? Second, in each such dimension, two equally weighted marks approaching one another attain this first-order rate. Which other arrangements of marks attain it? Third, our estimate fails in the plane. Already for two marks there, the fractional loss of inradius is asymptotically the square root of the fractional loss of area, which is much larger, as we saw after the proof of Theorem~\ref{thm:all-dimensions}. For a fixed number of equally weighted marks in the plane, what is the sharp bound on the fractional loss of inradius as the fractional loss of area tends to zero? \section*{Acknowledgments} This work was done overwhelmingly by AI models, steered by the author in a continuous critical dialogue, alternating between Anthropic's Claude and OpenAI's ChatGPT. The author takes full responsibility for the paper. \small \begin{thebibliography}{9}\setlength{\itemsep}{1pt} \bibitem{clark} C. Clark and R. S. Laugesen, \emph{Balls minimize moments of logarithmic and Newtonian equilibrium measures}, \href{https://doi.org/10.1007/s11118-024-10175-z}{Potential Anal. \textbf{63} (2025), 389--412}, published online in 2024. \bibitem{capacity} G. De Philippis, M. Marini and E. Mukoseeva, \emph{The sharp quantitative isocapacitary inequality}, \href{https://doi.org/10.4171/RMI/1259}{Rev. Mat. Iberoam. \textbf{37} (2021), 2191--2228}. \bibitem{erdos} P. Erd\H{o}s, \href{https://users.renyi.hu/~p_erdos/1976-12.pdf}{\emph{Extremal problems on polynomials}}, in \emph{Approximation Theory II}, Academic Press (1976), 347--355. \bibitem{short} D. Goel, \emph{Erd\H{o}s's 1974 distance-product volume problem resolved: the ball is always the unique maximizer in $d\ge2$}, preprint (2026).\par\nopagebreak \url{https://doi.org/10.5281/zenodo.24378921} \bibitem{hayman} W. K. Hayman and E. F. Lingham, \emph{Research Problems in Function Theory (New Edition)}, \href{https://arxiv.org/abs/1809.07200}{arXiv:1809.07200} (2018), Problem 4.23 and Tables 1--2. \bibitem{ferudun} A. Ferudun, \emph{On a question of Erd\H{o}s on lemniscates in $\R^m$: connected lemniscates lie in a ball of radius 2}, \href{https://doi.org/10.5281/zenodo.23422190}{preprint} (2026). \bibitem{talenti} G. Talenti, \href{https://www.numdam.org/item/ASNSP_1976_4_3_4_697_0/}{\emph{Elliptic equations and rearrangements}}, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) \textbf{3} (1976), 697--718. \bibitem{tang} H. Tang and Y. Wang, \emph{Sharp constants for weak estimates of the Riesz potentials}, \href{https://arxiv.org/abs/2608.31043v2}{arXiv:2608.31043v2} (2026). \bibitem{polarization} T. Erd\'elyi and E. B. Saff, \emph{Riesz polarization inequalities in higher dimensions}, \href{https://doi.org/10.1016/j.jat.2013.03.003}{J. Approx. Theory \textbf{171} (2013), 128--147}. \bibitem{jasion} P. Jasion and K. Magnucki, \href{https://am.ippt.gov.pl/index.php/am/article/view/v67p179}{\emph{Elastic buckling of Cassini ovaloidal shells under external pressure -- theoretical study}}, Arch. Mech. \textbf{67} (2015), 179--192. \bibitem{planar} P. Erd\H{o}s, F. Herzog and G. Piranian, \href{https://www.renyi.hu/~p_erdos/1958-05.pdf}{\emph{Metric properties of polynomials}}, J. Analyse Math. \textbf{6} (1958), 125--148. \bibitem{polya} G. P\'olya, \href{https://archive.org/details/sitzungsbericht1928preu}{\emph{Beitrag zur Verallgemeinerung des Verzerrungssatzes auf mehrfach zusammenh\"angende Gebiete (Zweite Mitteilung)}}, Sitzungsber. Preuss. Akad. Wiss., Phys.-Math. Kl. (1928), 280--282. \end{thebibliography} \end{document}